net/ipv4/fib_semantics.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              IPv4 Forwarding Information Base: semantics.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  */
  11
  12 #include <linux/uaccess.h>
  13 #include <linux/bitops.h>
  14 #include <linux/types.h>
  15 #include <linux/kernel.h>
  16 #include <linux/jiffies.h>
  17 #include <linux/mm.h>
  18 #include <linux/string.h>
  19 #include <linux/socket.h>
  20 #include <linux/sockios.h>
  21 #include <linux/errno.h>
  22 #include <linux/in.h>
  23 #include <linux/inet.h>
  24 #include <linux/inetdevice.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/if_arp.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/init.h>
  30 #include <linux/slab.h>
  31 #include <linux/netlink.h>
  32 #include <linux/hash.h>
  33 #include <linux/nospec.h>
  34
  35 #include <net/arp.h>
  36 #include <net/inet_dscp.h>
  37 #include <net/ip.h>
  38 #include <net/protocol.h>
  39 #include <net/route.h>
  40 #include <net/tcp.h>
  41 #include <net/sock.h>
  42 #include <net/ip_fib.h>
  43 #include <net/ip6_fib.h>
  44 #include <net/nexthop.h>
  45 #include <net/netlink.h>
  46 #include <net/rtnh.h>
  47 #include <net/lwtunnel.h>
  48 #include <net/fib_notifier.h>
  49 #include <net/addrconf.h>
  50
  51 #include "fib_lookup.h"
  52
  53 static DEFINE_SPINLOCK(fib_info_lock);
  54 static struct hlist_head *fib_info_hash;
  55 static struct hlist_head *fib_info_laddrhash;
  56 static unsigned int fib_info_hash_size;
  57 static unsigned int fib_info_hash_bits;
  58 static unsigned int fib_info_cnt;
  59
  60 #define DEVINDEX_HASHBITS 8
  61 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  62 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  63
  64 /* for_nexthops and change_nexthops only used when nexthop object
  65  * is not set in a fib_info. The logic within can reference fib_nh.
  66  */
  67 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  68
  69 #define for_nexthops(fi) {                                              \
  70         int nhsel; const struct fib_nh *nh;                             \
  71         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  72              nhsel < fib_info_num_path((fi));                           \
  73              nh++, nhsel++)
  74
  75 #define change_nexthops(fi) {                                           \
  76         int nhsel; struct fib_nh *nexthop_nh;                           \
  77         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  78              nhsel < fib_info_num_path((fi));                           \
  79              nexthop_nh++, nhsel++)
  80
  81 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  82
  83 /* Hope, that gcc will optimize it to get rid of dummy loop */
  84
  85 #define for_nexthops(fi) {                                              \
  86         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  87         for (nhsel = 0; nhsel < 1; nhsel++)
  88
  89 #define change_nexthops(fi) {                                           \
  90         int nhsel;                                                      \
  91         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  92         for (nhsel = 0; nhsel < 1; nhsel++)
  93
  94 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  95
  96 #define endfor_nexthops(fi) }
  97
  98
  99 const struct fib_prop fib_props[RTN_MAX + 1] = {
 100         [RTN_UNSPEC] = {
 101                 .error  = 0,
 102                 .scope  = RT_SCOPE_NOWHERE,
 103         },
 104         [RTN_UNICAST] = {
 105                 .error  = 0,
 106                 .scope  = RT_SCOPE_UNIVERSE,
 107         },
 108         [RTN_LOCAL] = {
 109                 .error  = 0,
 110                 .scope  = RT_SCOPE_HOST,
 111         },
 112         [RTN_BROADCAST] = {
 113                 .error  = 0,
 114                 .scope  = RT_SCOPE_LINK,
 115         },
 116         [RTN_ANYCAST] = {
 117                 .error  = 0,
 118                 .scope  = RT_SCOPE_LINK,
 119         },
 120         [RTN_MULTICAST] = {
 121                 .error  = 0,
 122                 .scope  = RT_SCOPE_UNIVERSE,
 123         },
 124         [RTN_BLACKHOLE] = {
 125                 .error  = -EINVAL,
 126                 .scope  = RT_SCOPE_UNIVERSE,
 127         },
 128         [RTN_UNREACHABLE] = {
 129                 .error  = -EHOSTUNREACH,
 130                 .scope  = RT_SCOPE_UNIVERSE,
 131         },
 132         [RTN_PROHIBIT] = {
 133                 .error  = -EACCES,
 134                 .scope  = RT_SCOPE_UNIVERSE,
 135         },
 136         [RTN_THROW] = {
 137                 .error  = -EAGAIN,
 138                 .scope  = RT_SCOPE_UNIVERSE,
 139         },
 140         [RTN_NAT] = {
 141                 .error  = -EINVAL,
 142                 .scope  = RT_SCOPE_NOWHERE,
 143         },
 144         [RTN_XRESOLVE] = {
 145                 .error  = -EINVAL,
 146                 .scope  = RT_SCOPE_NOWHERE,
 147         },
 148 };
 149
 150 static void rt_fibinfo_free(struct rtable __rcu **rtp)
 151 {
 152         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 153
 154         if (!rt)
 155                 return;
 156
 157         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 158          * because we waited an RCU grace period before calling
 159          * free_fib_info_rcu()
 160          */
 161
 162         dst_dev_put(&rt->dst);
 163         dst_release_immediate(&rt->dst);
 164 }
 165
 166 static void free_nh_exceptions(struct fib_nh_common *nhc)
 167 {
 168         struct fnhe_hash_bucket *hash;
 169         int i;
 170
 171         hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
 172         if (!hash)
 173                 return;
 174         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 175                 struct fib_nh_exception *fnhe;
 176
 177                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 178                 while (fnhe) {
 179                         struct fib_nh_exception *next;
 180
 181                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 182
 183                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 184                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 185
 186                         kfree(fnhe);
 187
 188                         fnhe = next;
 189                 }
 190         }
 191         kfree(hash);
 192 }
 193
 194 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 195 {
 196         int cpu;
 197
 198         if (!rtp)
 199                 return;
 200
 201         for_each_possible_cpu(cpu) {
 202                 struct rtable *rt;
 203
 204                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 205                 if (rt) {
 206                         dst_dev_put(&rt->dst);
 207                         dst_release_immediate(&rt->dst);
 208                 }
 209         }
 210         free_percpu(rtp);
 211 }
 212
 213 void fib_nh_common_release(struct fib_nh_common *nhc)
 214 {
 215         netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker);
 216         lwtstate_put(nhc->nhc_lwtstate);
 217         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 218         rt_fibinfo_free(&nhc->nhc_rth_input);
 219         free_nh_exceptions(nhc);
 220 }
 221 EXPORT_SYMBOL_GPL(fib_nh_common_release);
 222
 223 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
 224 {
 225 #ifdef CONFIG_IP_ROUTE_CLASSID
 226         if (fib_nh->nh_tclassid)
 227                 atomic_dec(&net->ipv4.fib_num_tclassid_users);
 228 #endif
 229         fib_nh_common_release(&fib_nh->nh_common);
 230 }
 231
 232 /* Release a nexthop info record */
 233 static void free_fib_info_rcu(struct rcu_head *head)
 234 {
 235         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 236
 237         if (fi->nh) {
 238                 nexthop_put(fi->nh);
 239         } else {
 240                 change_nexthops(fi) {
 241                         fib_nh_release(fi->fib_net, nexthop_nh);
 242                 } endfor_nexthops(fi);
 243         }
 244
 245         ip_fib_metrics_put(fi->fib_metrics);
 246
 247         kfree(fi);
 248 }
 249
 250 void free_fib_info(struct fib_info *fi)
 251 {
 252         if (fi->fib_dead == 0) {
 253                 pr_warn("Freeing alive fib_info %p\n", fi);
 254                 return;
 255         }
 256
 257         call_rcu(&fi->rcu, free_fib_info_rcu);
 258 }
 259 EXPORT_SYMBOL_GPL(free_fib_info);
 260
 261 void fib_release_info(struct fib_info *fi)
 262 {
 263         spin_lock_bh(&fib_info_lock);
 264         if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
 265                 hlist_del(&fi->fib_hash);
 266
 267                 /* Paired with READ_ONCE() in fib_create_info(). */
 268                 WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
 269
 270                 if (fi->fib_prefsrc)
 271                         hlist_del(&fi->fib_lhash);
 272                 if (fi->nh) {
 273                         list_del(&fi->nh_list);
 274                 } else {
 275                         change_nexthops(fi) {
 276                                 if (!nexthop_nh->fib_nh_dev)
 277                                         continue;
 278                                 hlist_del(&nexthop_nh->nh_hash);
 279                         } endfor_nexthops(fi)
 280                 }
 281                 fi->fib_dead = 1;
 282                 fib_info_put(fi);
 283         }
 284         spin_unlock_bh(&fib_info_lock);
 285 }
 286
 287 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
 288 {
 289         const struct fib_nh *onh;
 290
 291         if (fi->nh || ofi->nh)
 292                 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
 293
 294         if (ofi->fib_nhs == 0)
 295                 return 0;
 296
 297         for_nexthops(fi) {
 298                 onh = fib_info_nh(ofi, nhsel);
 299
 300                 if (nh->fib_nh_oif != onh->fib_nh_oif ||
 301                     nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
 302                     nh->fib_nh_scope != onh->fib_nh_scope ||
 303 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 304                     nh->fib_nh_weight != onh->fib_nh_weight ||
 305 #endif
 306 #ifdef CONFIG_IP_ROUTE_CLASSID
 307                     nh->nh_tclassid != onh->nh_tclassid ||
 308 #endif
 309                     lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
 310                     ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
 311                         return -1;
 312
 313                 if (nh->fib_nh_gw_family == AF_INET &&
 314                     nh->fib_nh_gw4 != onh->fib_nh_gw4)
 315                         return -1;
 316
 317                 if (nh->fib_nh_gw_family == AF_INET6 &&
 318                     ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
 319                         return -1;
 320         } endfor_nexthops(fi);
 321         return 0;
 322 }
 323
 324 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 325 {
 326         return hash_32(val, DEVINDEX_HASHBITS);
 327 }
 328
 329 static struct hlist_head *
 330 fib_info_devhash_bucket(const struct net_device *dev)
 331 {
 332         u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
 333
 334         return &fib_info_devhash[fib_devindex_hashfn(val)];
 335 }
 336
 337 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
 338                                       u32 prefsrc, u32 priority)
 339 {
 340         unsigned int val = init_val;
 341
 342         val ^= (protocol << 8) | scope;
 343         val ^= prefsrc;
 344         val ^= priority;
 345
 346         return val;
 347 }
 348
 349 static unsigned int fib_info_hashfn_result(unsigned int val)
 350 {
 351         unsigned int mask = (fib_info_hash_size - 1);
 352
 353         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 354 }
 355
 356 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
 357 {
 358         unsigned int val;
 359
 360         val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
 361                                 fi->fib_scope, (__force u32)fi->fib_prefsrc,
 362                                 fi->fib_priority);
 363
 364         if (fi->nh) {
 365                 val ^= fib_devindex_hashfn(fi->nh->id);
 366         } else {
 367                 for_nexthops(fi) {
 368                         val ^= fib_devindex_hashfn(nh->fib_nh_oif);
 369                 } endfor_nexthops(fi)
 370         }
 371
 372         return fib_info_hashfn_result(val);
 373 }
 374
 375 /* no metrics, only nexthop id */
 376 static struct fib_info *fib_find_info_nh(struct net *net,
 377                                          const struct fib_config *cfg)
 378 {
 379         struct hlist_head *head;
 380         struct fib_info *fi;
 381         unsigned int hash;
 382
 383         hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
 384                                  cfg->fc_protocol, cfg->fc_scope,
 385                                  (__force u32)cfg->fc_prefsrc,
 386                                  cfg->fc_priority);
 387         hash = fib_info_hashfn_result(hash);
 388         head = &fib_info_hash[hash];
 389
 390         hlist_for_each_entry(fi, head, fib_hash) {
 391                 if (!net_eq(fi->fib_net, net))
 392                         continue;
 393                 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
 394                         continue;
 395                 if (cfg->fc_protocol == fi->fib_protocol &&
 396                     cfg->fc_scope == fi->fib_scope &&
 397                     cfg->fc_prefsrc == fi->fib_prefsrc &&
 398                     cfg->fc_priority == fi->fib_priority &&
 399                     cfg->fc_type == fi->fib_type &&
 400                     cfg->fc_table == fi->fib_tb_id &&
 401                     !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
 402                         return fi;
 403         }
 404
 405         return NULL;
 406 }
 407
 408 static struct fib_info *fib_find_info(struct fib_info *nfi)
 409 {
 410         struct hlist_head *head;
 411         struct fib_info *fi;
 412         unsigned int hash;
 413
 414         hash = fib_info_hashfn(nfi);
 415         head = &fib_info_hash[hash];
 416
 417         hlist_for_each_entry(fi, head, fib_hash) {
 418                 if (!net_eq(fi->fib_net, nfi->fib_net))
 419                         continue;
 420                 if (fi->fib_nhs != nfi->fib_nhs)
 421                         continue;
 422                 if (nfi->fib_protocol == fi->fib_protocol &&
 423                     nfi->fib_scope == fi->fib_scope &&
 424                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 425                     nfi->fib_priority == fi->fib_priority &&
 426                     nfi->fib_type == fi->fib_type &&
 427                     nfi->fib_tb_id == fi->fib_tb_id &&
 428                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 429                            sizeof(u32) * RTAX_MAX) == 0 &&
 430                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
 431                     nh_comp(fi, nfi) == 0)
 432                         return fi;
 433         }
 434
 435         return NULL;
 436 }
 437
 438 /* Check, that the gateway is already configured.
 439  * Used only by redirect accept routine.
 440  */
 441 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 442 {
 443         struct hlist_head *head;
 444         struct fib_nh *nh;
 445
 446         spin_lock(&fib_info_lock);
 447
 448         head = fib_info_devhash_bucket(dev);
 449
 450         hlist_for_each_entry(nh, head, nh_hash) {
 451                 if (nh->fib_nh_dev == dev &&
 452                     nh->fib_nh_gw4 == gw &&
 453                     !(nh->fib_nh_flags & RTNH_F_DEAD)) {
 454                         spin_unlock(&fib_info_lock);
 455                         return 0;
 456                 }
 457         }
 458
 459         spin_unlock(&fib_info_lock);
 460
 461         return -1;
 462 }
 463
 464 size_t fib_nlmsg_size(struct fib_info *fi)
 465 {
 466         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 467                          + nla_total_size(4) /* RTA_TABLE */
 468                          + nla_total_size(4) /* RTA_DST */
 469                          + nla_total_size(4) /* RTA_PRIORITY */
 470                          + nla_total_size(4) /* RTA_PREFSRC */
 471                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 472         unsigned int nhs = fib_info_num_path(fi);
 473
 474         /* space for nested metrics */
 475         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 476
 477         if (fi->nh)
 478                 payload += nla_total_size(4); /* RTA_NH_ID */
 479
 480         if (nhs) {
 481                 size_t nh_encapsize = 0;
 482                 /* Also handles the special case nhs == 1 */
 483
 484                 /* each nexthop is packed in an attribute */
 485                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 486                 unsigned int i;
 487
 488                 /* may contain flow and gateway attribute */
 489                 nhsize += 2 * nla_total_size(4);
 490
 491                 /* grab encap info */
 492                 for (i = 0; i < fib_info_num_path(fi); i++) {
 493                         struct fib_nh_common *nhc = fib_info_nhc(fi, i);
 494
 495                         if (nhc->nhc_lwtstate) {
 496                                 /* RTA_ENCAP_TYPE */
 497                                 nh_encapsize += lwtunnel_get_encap_size(
 498                                                 nhc->nhc_lwtstate);
 499                                 /* RTA_ENCAP */
 500                                 nh_encapsize +=  nla_total_size(2);
 501                         }
 502                 }
 503
 504                 /* all nexthops are packed in a nested attribute */
 505                 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
 506
 507         }
 508
 509         return payload;
 510 }
 511
 512 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 513                int dst_len, u32 tb_id, const struct nl_info *info,
 514                unsigned int nlm_flags)
 515 {
 516         struct fib_rt_info fri;
 517         struct sk_buff *skb;
 518         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 519         int err = -ENOBUFS;
 520
 521         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 522         if (!skb)
 523                 goto errout;
 524
 525         fri.fi = fa->fa_info;
 526         fri.tb_id = tb_id;
 527         fri.dst = key;
 528         fri.dst_len = dst_len;
 529         fri.dscp = fa->fa_dscp;
 530         fri.type = fa->fa_type;
 531         fri.offload = READ_ONCE(fa->offload);
 532         fri.trap = READ_ONCE(fa->trap);
 533         fri.offload_failed = READ_ONCE(fa->offload_failed);
 534         err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
 535         if (err < 0) {
 536                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 537                 WARN_ON(err == -EMSGSIZE);
 538                 kfree_skb(skb);
 539                 goto errout;
 540         }
 541         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 542                     info->nlh, GFP_KERNEL);
 543         return;
 544 errout:
 545         if (err < 0)
 546                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 547 }
 548
 549 static int fib_detect_death(struct fib_info *fi, int order,
 550                             struct fib_info **last_resort, int *last_idx,
 551                             int dflt)
 552 {
 553         const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
 554         struct neighbour *n;
 555         int state = NUD_NONE;
 556
 557         if (likely(nhc->nhc_gw_family == AF_INET))
 558                 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
 559         else if (nhc->nhc_gw_family == AF_INET6)
 560                 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
 561                                  nhc->nhc_dev);
 562         else
 563                 n = NULL;
 564
 565         if (n) {
 566                 state = READ_ONCE(n->nud_state);
 567                 neigh_release(n);
 568         } else {
 569                 return 0;
 570         }
 571         if (state == NUD_REACHABLE)
 572                 return 0;
 573         if ((state & NUD_VALID) && order != dflt)
 574                 return 0;
 575         if ((state & NUD_VALID) ||
 576             (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
 577                 *last_resort = fi;
 578                 *last_idx = order;
 579         }
 580         return 1;
 581 }
 582
 583 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
 584                        struct nlattr *encap, u16 encap_type,
 585                        void *cfg, gfp_t gfp_flags,
 586                        struct netlink_ext_ack *extack)
 587 {
 588         int err;
 589
 590         nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
 591                                                     gfp_flags);
 592         if (!nhc->nhc_pcpu_rth_output)
 593                 return -ENOMEM;
 594
 595         if (encap) {
 596                 struct lwtunnel_state *lwtstate;
 597
 598                 if (encap_type == LWTUNNEL_ENCAP_NONE) {
 599                         NL_SET_ERR_MSG(extack, "LWT encap type not specified");
 600                         err = -EINVAL;
 601                         goto lwt_failure;
 602                 }
 603                 err = lwtunnel_build_state(net, encap_type, encap,
 604                                            nhc->nhc_family, cfg, &lwtstate,
 605                                            extack);
 606                 if (err)
 607                         goto lwt_failure;
 608
 609                 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
 610         }
 611
 612         return 0;
 613
 614 lwt_failure:
 615         rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
 616         nhc->nhc_pcpu_rth_output = NULL;
 617         return err;
 618 }
 619 EXPORT_SYMBOL_GPL(fib_nh_common_init);
 620
 621 int fib_nh_init(struct net *net, struct fib_nh *nh,
 622                 struct fib_config *cfg, int nh_weight,
 623                 struct netlink_ext_ack *extack)
 624 {
 625         int err;
 626
 627         nh->fib_nh_family = AF_INET;
 628
 629         err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
 630                                  cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
 631         if (err)
 632                 return err;
 633
 634         nh->fib_nh_oif = cfg->fc_oif;
 635         nh->fib_nh_gw_family = cfg->fc_gw_family;
 636         if (cfg->fc_gw_family == AF_INET)
 637                 nh->fib_nh_gw4 = cfg->fc_gw4;
 638         else if (cfg->fc_gw_family == AF_INET6)
 639                 nh->fib_nh_gw6 = cfg->fc_gw6;
 640
 641         nh->fib_nh_flags = cfg->fc_flags;
 642
 643 #ifdef CONFIG_IP_ROUTE_CLASSID
 644         nh->nh_tclassid = cfg->fc_flow;
 645         if (nh->nh_tclassid)
 646                 atomic_inc(&net->ipv4.fib_num_tclassid_users);
 647 #endif
 648 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 649         nh->fib_nh_weight = nh_weight;
 650 #endif
 651         return 0;
 652 }
 653
 654 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 655
 656 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
 657                               struct netlink_ext_ack *extack)
 658 {
 659         int nhs = 0;
 660
 661         while (rtnh_ok(rtnh, remaining)) {
 662                 nhs++;
 663                 rtnh = rtnh_next(rtnh, &remaining);
 664         }
 665
 666         /* leftover implies invalid nexthop configuration, discard it */
 667         if (remaining > 0) {
 668                 NL_SET_ERR_MSG(extack,
 669                                "Invalid nexthop configuration - extra data after nexthops");
 670                 nhs = 0;
 671         }
 672
 673         return nhs;
 674 }
 675
 676 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
 677                             struct netlink_ext_ack *extack)
 678 {
 679         if (nla_len(nla) < sizeof(*gw)) {
 680                 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
 681                 return -EINVAL;
 682         }
 683
 684         *gw = nla_get_in_addr(nla);
 685
 686         return 0;
 687 }
 688
 689 /* only called when fib_nh is integrated into fib_info */
 690 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 691                        int remaining, struct fib_config *cfg,
 692                        struct netlink_ext_ack *extack)
 693 {
 694         struct net *net = fi->fib_net;
 695         struct fib_config fib_cfg;
 696         struct fib_nh *nh;
 697         int ret;
 698
 699         change_nexthops(fi) {
 700                 int attrlen;
 701
 702                 memset(&fib_cfg, 0, sizeof(fib_cfg));
 703
 704                 if (!rtnh_ok(rtnh, remaining)) {
 705                         NL_SET_ERR_MSG(extack,
 706                                        "Invalid nexthop configuration - extra data after nexthop");
 707                         return -EINVAL;
 708                 }
 709
 710                 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
 711                         NL_SET_ERR_MSG(extack,
 712                                        "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
 713                         return -EINVAL;
 714                 }
 715
 716                 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 717                 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
 718
 719                 attrlen = rtnh_attrlen(rtnh);
 720                 if (attrlen > 0) {
 721                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 722
 723                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 724                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 725                         if (nla && nlav) {
 726                                 NL_SET_ERR_MSG(extack,
 727                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 728                                 return -EINVAL;
 729                         }
 730                         if (nla) {
 731                                 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
 732                                                        extack);
 733                                 if (ret)
 734                                         goto errout;
 735
 736                                 if (fib_cfg.fc_gw4)
 737                                         fib_cfg.fc_gw_family = AF_INET;
 738                         } else if (nlav) {
 739                                 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
 740                                 if (ret)
 741                                         goto errout;
 742                         }
 743
 744                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 745                         if (nla) {
 746                                 if (nla_len(nla) < sizeof(u32)) {
 747                                         NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
 748                                         return -EINVAL;
 749                                 }
 750                                 fib_cfg.fc_flow = nla_get_u32(nla);
 751                         }
 752
 753                         fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
 754                         /* RTA_ENCAP_TYPE length checked in
 755                          * lwtunnel_valid_encap_type_attr
 756                          */
 757                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
 758                         if (nla)
 759                                 fib_cfg.fc_encap_type = nla_get_u16(nla);
 760                 }
 761
 762                 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
 763                                   rtnh->rtnh_hops + 1, extack);
 764                 if (ret)
 765                         goto errout;
 766
 767                 rtnh = rtnh_next(rtnh, &remaining);
 768         } endfor_nexthops(fi);
 769
 770         ret = -EINVAL;
 771         nh = fib_info_nh(fi, 0);
 772         if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
 773                 NL_SET_ERR_MSG(extack,
 774                                "Nexthop device index does not match RTA_OIF");
 775                 goto errout;
 776         }
 777         if (cfg->fc_gw_family) {
 778                 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
 779                     (cfg->fc_gw_family == AF_INET &&
 780                      nh->fib_nh_gw4 != cfg->fc_gw4) ||
 781                     (cfg->fc_gw_family == AF_INET6 &&
 782                      ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
 783                         NL_SET_ERR_MSG(extack,
 784                                        "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
 785                         goto errout;
 786                 }
 787         }
 788 #ifdef CONFIG_IP_ROUTE_CLASSID
 789         if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
 790                 NL_SET_ERR_MSG(extack,
 791                                "Nexthop class id does not match RTA_FLOW");
 792                 goto errout;
 793         }
 794 #endif
 795         ret = 0;
 796 errout:
 797         return ret;
 798 }
 799
 800 /* only called when fib_nh is integrated into fib_info */
 801 static void fib_rebalance(struct fib_info *fi)
 802 {
 803         int total;
 804         int w;
 805
 806         if (fib_info_num_path(fi) < 2)
 807                 return;
 808
 809         total = 0;
 810         for_nexthops(fi) {
 811                 if (nh->fib_nh_flags & RTNH_F_DEAD)
 812                         continue;
 813
 814                 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
 815                     nh->fib_nh_flags & RTNH_F_LINKDOWN)
 816                         continue;
 817
 818                 total += nh->fib_nh_weight;
 819         } endfor_nexthops(fi);
 820
 821         w = 0;
 822         change_nexthops(fi) {
 823                 int upper_bound;
 824
 825                 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
 826                         upper_bound = -1;
 827                 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
 828                            nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
 829                         upper_bound = -1;
 830                 } else {
 831                         w += nexthop_nh->fib_nh_weight;
 832                         upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
 833                                                             total) - 1;
 834                 }
 835
 836                 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
 837         } endfor_nexthops(fi);
 838 }
 839 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 840
 841 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 842                        int remaining, struct fib_config *cfg,
 843                        struct netlink_ext_ack *extack)
 844 {
 845         NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
 846
 847         return -EINVAL;
 848 }
 849
 850 #define fib_rebalance(fi) do { } while (0)
 851
 852 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 853
 854 static int fib_encap_match(struct net *net, u16 encap_type,
 855                            struct nlattr *encap,
 856                            const struct fib_nh *nh,
 857                            const struct fib_config *cfg,
 858                            struct netlink_ext_ack *extack)
 859 {
 860         struct lwtunnel_state *lwtstate;
 861         int ret, result = 0;
 862
 863         if (encap_type == LWTUNNEL_ENCAP_NONE)
 864                 return 0;
 865
 866         ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
 867                                    cfg, &lwtstate, extack);
 868         if (!ret) {
 869                 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
 870                 lwtstate_free(lwtstate);
 871         }
 872
 873         return result;
 874 }
 875
 876 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
 877                  struct netlink_ext_ack *extack)
 878 {
 879 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 880         struct rtnexthop *rtnh;
 881         int remaining;
 882 #endif
 883
 884         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 885                 return 1;
 886
 887         if (cfg->fc_nh_id) {
 888                 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
 889                         return 0;
 890                 return 1;
 891         }
 892
 893         if (fi->nh) {
 894                 if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
 895                         return 1;
 896                 return 0;
 897         }
 898
 899         if (cfg->fc_oif || cfg->fc_gw_family) {
 900                 struct fib_nh *nh;
 901
 902                 nh = fib_info_nh(fi, 0);
 903                 if (cfg->fc_encap) {
 904                         if (fib_encap_match(net, cfg->fc_encap_type,
 905                                             cfg->fc_encap, nh, cfg, extack))
 906                                 return 1;
 907                 }
 908 #ifdef CONFIG_IP_ROUTE_CLASSID
 909                 if (cfg->fc_flow &&
 910                     cfg->fc_flow != nh->nh_tclassid)
 911                         return 1;
 912 #endif
 913                 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
 914                     (cfg->fc_gw_family &&
 915                      cfg->fc_gw_family != nh->fib_nh_gw_family))
 916                         return 1;
 917
 918                 if (cfg->fc_gw_family == AF_INET &&
 919                     cfg->fc_gw4 != nh->fib_nh_gw4)
 920                         return 1;
 921
 922                 if (cfg->fc_gw_family == AF_INET6 &&
 923                     ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
 924                         return 1;
 925
 926                 return 0;
 927         }
 928
 929 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 930         if (!cfg->fc_mp)
 931                 return 0;
 932
 933         rtnh = cfg->fc_mp;
 934         remaining = cfg->fc_mp_len;
 935
 936         for_nexthops(fi) {
 937                 int attrlen;
 938
 939                 if (!rtnh_ok(rtnh, remaining))
 940                         return -EINVAL;
 941
 942                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
 943                         return 1;
 944
 945                 attrlen = rtnh_attrlen(rtnh);
 946                 if (attrlen > 0) {
 947                         struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
 948                         int err;
 949
 950                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 951                         nlav = nla_find(attrs, attrlen, RTA_VIA);
 952                         if (nla && nlav) {
 953                                 NL_SET_ERR_MSG(extack,
 954                                                "Nexthop configuration can not contain both GATEWAY and VIA");
 955                                 return -EINVAL;
 956                         }
 957
 958                         if (nla) {
 959                                 __be32 gw;
 960
 961                                 err = fib_gw_from_attr(&gw, nla, extack);
 962                                 if (err)
 963                                         return err;
 964
 965                                 if (nh->fib_nh_gw_family != AF_INET ||
 966                                     gw != nh->fib_nh_gw4)
 967                                         return 1;
 968                         } else if (nlav) {
 969                                 struct fib_config cfg2;
 970
 971                                 err = fib_gw_from_via(&cfg2, nlav, extack);
 972                                 if (err)
 973                                         return err;
 974
 975                                 switch (nh->fib_nh_gw_family) {
 976                                 case AF_INET:
 977                                         if (cfg2.fc_gw_family != AF_INET ||
 978                                             cfg2.fc_gw4 != nh->fib_nh_gw4)
 979                                                 return 1;
 980                                         break;
 981                                 case AF_INET6:
 982                                         if (cfg2.fc_gw_family != AF_INET6 ||
 983                                             ipv6_addr_cmp(&cfg2.fc_gw6,
 984                                                           &nh->fib_nh_gw6))
 985                                                 return 1;
 986                                         break;
 987                                 }
 988                         }
 989
 990 #ifdef CONFIG_IP_ROUTE_CLASSID
 991                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 992                         if (nla) {
 993                                 if (nla_len(nla) < sizeof(u32)) {
 994                                         NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
 995                                         return -EINVAL;
 996                                 }
 997                                 if (nla_get_u32(nla) != nh->nh_tclassid)
 998                                         return 1;
 999                         }
1000 #endif
1001                 }
1002
1003                 rtnh = rtnh_next(rtnh, &remaining);
1004         } endfor_nexthops(fi);
1005 #endif
1006         return 0;
1007 }
1008
1009 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1010 {
1011         struct nlattr *nla;
1012         int remaining;
1013
1014         if (!cfg->fc_mx)
1015                 return true;
1016
1017         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1018                 int type = nla_type(nla);
1019                 u32 fi_val, val;
1020
1021                 if (!type)
1022                         continue;
1023                 if (type > RTAX_MAX)
1024                         return false;
1025
1026                 type = array_index_nospec(type, RTAX_MAX + 1);
1027                 if (type == RTAX_CC_ALGO) {
1028                         char tmp[TCP_CA_NAME_MAX];
1029                         bool ecn_ca = false;
1030
1031                         nla_strscpy(tmp, nla, sizeof(tmp));
1032                         val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1033                 } else {
1034                         if (nla_len(nla) != sizeof(u32))
1035                                 return false;
1036                         val = nla_get_u32(nla);
1037                 }
1038
1039                 fi_val = fi->fib_metrics->metrics[type - 1];
1040                 if (type == RTAX_FEATURES)
1041                         fi_val &= ~DST_FEATURE_ECN_CA;
1042
1043                 if (fi_val != val)
1044                         return false;
1045         }
1046
1047         return true;
1048 }
1049
1050 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1051                               u32 table, struct netlink_ext_ack *extack)
1052 {
1053         struct fib6_config cfg = {
1054                 .fc_table = table,
1055                 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1056                 .fc_ifindex = nh->fib_nh_oif,
1057                 .fc_gateway = nh->fib_nh_gw6,
1058         };
1059         struct fib6_nh fib6_nh = {};
1060         int err;
1061
1062         err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1063         if (!err) {
1064                 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1065                 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1066                             GFP_KERNEL);
1067                 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1068                 nh->fib_nh_scope = RT_SCOPE_LINK;
1069
1070                 ipv6_stub->fib6_nh_release(&fib6_nh);
1071         }
1072
1073         return err;
1074 }
1075
1076 /*
1077  * Picture
1078  * -------
1079  *
1080  * Semantics of nexthop is very messy by historical reasons.
1081  * We have to take into account, that:
1082  * a) gateway can be actually local interface address,
1083  *    so that gatewayed route is direct.
1084  * b) gateway must be on-link address, possibly
1085  *    described not by an ifaddr, but also by a direct route.
1086  * c) If both gateway and interface are specified, they should not
1087  *    contradict.
1088  * d) If we use tunnel routes, gateway could be not on-link.
1089  *
1090  * Attempt to reconcile all of these (alas, self-contradictory) conditions
1091  * results in pretty ugly and hairy code with obscure logic.
1092  *
1093  * I chose to generalized it instead, so that the size
1094  * of code does not increase practically, but it becomes
1095  * much more general.
1096  * Every prefix is assigned a "scope" value: "host" is local address,
1097  * "link" is direct route,
1098  * [ ... "site" ... "interior" ... ]
1099  * and "universe" is true gateway route with global meaning.
1100  *
1101  * Every prefix refers to a set of "nexthop"s (gw, oif),
1102  * where gw must have narrower scope. This recursion stops
1103  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1104  * which means that gw is forced to be on link.
1105  *
1106  * Code is still hairy, but now it is apparently logically
1107  * consistent and very flexible. F.e. as by-product it allows
1108  * to co-exists in peace independent exterior and interior
1109  * routing processes.
1110  *
1111  * Normally it looks as following.
1112  *
1113  * {universe prefix}  -> (gw, oif) [scope link]
1114  *                |
1115  *                |-> {link prefix} -> (gw, oif) [scope local]
1116  *                                      |
1117  *                                      |-> {local prefix} (terminal node)
1118  */
1119 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1120                               u8 scope, struct netlink_ext_ack *extack)
1121 {
1122         struct net_device *dev;
1123         struct fib_result res;
1124         int err = 0;
1125
1126         if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1127                 unsigned int addr_type;
1128
1129                 if (scope >= RT_SCOPE_LINK) {
1130                         NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1131                         return -EINVAL;
1132                 }
1133                 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1134                 if (!dev) {
1135                         NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1136                         return -ENODEV;
1137                 }
1138                 if (!(dev->flags & IFF_UP)) {
1139                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1140                         return -ENETDOWN;
1141                 }
1142                 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1143                 if (addr_type != RTN_UNICAST) {
1144                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1145                         return -EINVAL;
1146                 }
1147                 if (!netif_carrier_ok(dev))
1148                         nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1149                 nh->fib_nh_dev = dev;
1150                 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1151                 nh->fib_nh_scope = RT_SCOPE_LINK;
1152                 return 0;
1153         }
1154         rcu_read_lock();
1155         {
1156                 struct fib_table *tbl = NULL;
1157                 struct flowi4 fl4 = {
1158                         .daddr = nh->fib_nh_gw4,
1159                         .flowi4_scope = scope + 1,
1160                         .flowi4_oif = nh->fib_nh_oif,
1161                         .flowi4_iif = LOOPBACK_IFINDEX,
1162                 };
1163
1164                 /* It is not necessary, but requires a bit of thinking */
1165                 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1166                         fl4.flowi4_scope = RT_SCOPE_LINK;
1167
1168                 if (table && table != RT_TABLE_MAIN)
1169                         tbl = fib_get_table(net, table);
1170
1171                 if (tbl)
1172                         err = fib_table_lookup(tbl, &fl4, &res,
1173                                                FIB_LOOKUP_IGNORE_LINKSTATE |
1174                                                FIB_LOOKUP_NOREF);
1175
1176                 /* on error or if no table given do full lookup. This
1177                  * is needed for example when nexthops are in the local
1178                  * table rather than the given table
1179                  */
1180                 if (!tbl || err) {
1181                         err = fib_lookup(net, &fl4, &res,
1182                                          FIB_LOOKUP_IGNORE_LINKSTATE);
1183                 }
1184
1185                 if (err) {
1186                         NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1187                         goto out;
1188                 }
1189         }
1190
1191         err = -EINVAL;
1192         if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1193                 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1194                 goto out;
1195         }
1196         nh->fib_nh_scope = res.scope;
1197         nh->fib_nh_oif = FIB_RES_OIF(res);
1198         nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1199         if (!dev) {
1200                 NL_SET_ERR_MSG(extack,
1201                                "No egress device for nexthop gateway");
1202                 goto out;
1203         }
1204         netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1205         if (!netif_carrier_ok(dev))
1206                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1207         err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1208 out:
1209         rcu_read_unlock();
1210         return err;
1211 }
1212
1213 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1214                               struct netlink_ext_ack *extack)
1215 {
1216         struct in_device *in_dev;
1217         int err;
1218
1219         if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1220                 NL_SET_ERR_MSG(extack,
1221                                "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1222                 return -EINVAL;
1223         }
1224
1225         rcu_read_lock();
1226
1227         err = -ENODEV;
1228         in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1229         if (!in_dev)
1230                 goto out;
1231         err = -ENETDOWN;
1232         if (!(in_dev->dev->flags & IFF_UP)) {
1233                 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1234                 goto out;
1235         }
1236
1237         nh->fib_nh_dev = in_dev->dev;
1238         netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1239         nh->fib_nh_scope = RT_SCOPE_HOST;
1240         if (!netif_carrier_ok(nh->fib_nh_dev))
1241                 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1242         err = 0;
1243 out:
1244         rcu_read_unlock();
1245         return err;
1246 }
1247
1248 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1249                  struct netlink_ext_ack *extack)
1250 {
1251         int err;
1252
1253         if (nh->fib_nh_gw_family == AF_INET)
1254                 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1255         else if (nh->fib_nh_gw_family == AF_INET6)
1256                 err = fib_check_nh_v6_gw(net, nh, table, extack);
1257         else
1258                 err = fib_check_nh_nongw(net, nh, extack);
1259
1260         return err;
1261 }
1262
1263 static struct hlist_head *
1264 fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1265 {
1266         u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1267                            fib_info_hash_bits);
1268
1269         return &fib_info_laddrhash[slot];
1270 }
1271
1272 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1273                                struct hlist_head *new_laddrhash,
1274                                unsigned int new_size)
1275 {
1276         struct hlist_head *old_info_hash, *old_laddrhash;
1277         unsigned int old_size = fib_info_hash_size;
1278         unsigned int i;
1279
1280         spin_lock_bh(&fib_info_lock);
1281         old_info_hash = fib_info_hash;
1282         old_laddrhash = fib_info_laddrhash;
1283         fib_info_hash_size = new_size;
1284         fib_info_hash_bits = ilog2(new_size);
1285
1286         for (i = 0; i < old_size; i++) {
1287                 struct hlist_head *head = &fib_info_hash[i];
1288                 struct hlist_node *n;
1289                 struct fib_info *fi;
1290
1291                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1292                         struct hlist_head *dest;
1293                         unsigned int new_hash;
1294
1295                         new_hash = fib_info_hashfn(fi);
1296                         dest = &new_info_hash[new_hash];
1297                         hlist_add_head(&fi->fib_hash, dest);
1298                 }
1299         }
1300         fib_info_hash = new_info_hash;
1301
1302         fib_info_laddrhash = new_laddrhash;
1303         for (i = 0; i < old_size; i++) {
1304                 struct hlist_head *lhead = &old_laddrhash[i];
1305                 struct hlist_node *n;
1306                 struct fib_info *fi;
1307
1308                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1309                         struct hlist_head *ldest;
1310
1311                         ldest = fib_info_laddrhash_bucket(fi->fib_net,
1312                                                           fi->fib_prefsrc);
1313                         hlist_add_head(&fi->fib_lhash, ldest);
1314                 }
1315         }
1316
1317         spin_unlock_bh(&fib_info_lock);
1318
1319         kvfree(old_info_hash);
1320         kvfree(old_laddrhash);
1321 }
1322
1323 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1324                                  unsigned char scope)
1325 {
1326         struct fib_nh *nh;
1327
1328         if (nhc->nhc_family != AF_INET)
1329                 return inet_select_addr(nhc->nhc_dev, 0, scope);
1330
1331         nh = container_of(nhc, struct fib_nh, nh_common);
1332         nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1333         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1334
1335         return nh->nh_saddr;
1336 }
1337
1338 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1339 {
1340         struct fib_nh_common *nhc = res->nhc;
1341
1342         if (res->fi->fib_prefsrc)
1343                 return res->fi->fib_prefsrc;
1344
1345         if (nhc->nhc_family == AF_INET) {
1346                 struct fib_nh *nh;
1347
1348                 nh = container_of(nhc, struct fib_nh, nh_common);
1349                 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1350                         return nh->nh_saddr;
1351         }
1352
1353         return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1354 }
1355
1356 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1357 {
1358         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1359             fib_prefsrc != cfg->fc_dst) {
1360                 u32 tb_id = cfg->fc_table;
1361                 int rc;
1362
1363                 if (tb_id == RT_TABLE_MAIN)
1364                         tb_id = RT_TABLE_LOCAL;
1365
1366                 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1367                                           fib_prefsrc, tb_id);
1368
1369                 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1370                         rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1371                                                   fib_prefsrc, RT_TABLE_LOCAL);
1372                 }
1373
1374                 if (rc != RTN_LOCAL)
1375                         return false;
1376         }
1377         return true;
1378 }
1379
1380 struct fib_info *fib_create_info(struct fib_config *cfg,
1381                                  struct netlink_ext_ack *extack)
1382 {
1383         int err;
1384         struct fib_info *fi = NULL;
1385         struct nexthop *nh = NULL;
1386         struct fib_info *ofi;
1387         int nhs = 1;
1388         struct net *net = cfg->fc_nlinfo.nl_net;
1389
1390         if (cfg->fc_type > RTN_MAX)
1391                 goto err_inval;
1392
1393         /* Fast check to catch the most weird cases */
1394         if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1395                 NL_SET_ERR_MSG(extack, "Invalid scope");
1396                 goto err_inval;
1397         }
1398
1399         if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1400                 NL_SET_ERR_MSG(extack,
1401                                "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1402                 goto err_inval;
1403         }
1404
1405         if (cfg->fc_nh_id) {
1406                 if (!cfg->fc_mx) {
1407                         fi = fib_find_info_nh(net, cfg);
1408                         if (fi) {
1409                                 refcount_inc(&fi->fib_treeref);
1410                                 return fi;
1411                         }
1412                 }
1413
1414                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1415                 if (!nh) {
1416                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1417                         goto err_inval;
1418                 }
1419                 nhs = 0;
1420         }
1421
1422 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1423         if (cfg->fc_mp) {
1424                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1425                 if (nhs == 0)
1426                         goto err_inval;
1427         }
1428 #endif
1429
1430         err = -ENOBUFS;
1431
1432         /* Paired with WRITE_ONCE() in fib_release_info() */
1433         if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1434                 unsigned int new_size = fib_info_hash_size << 1;
1435                 struct hlist_head *new_info_hash;
1436                 struct hlist_head *new_laddrhash;
1437                 size_t bytes;
1438
1439                 if (!new_size)
1440                         new_size = 16;
1441                 bytes = (size_t)new_size * sizeof(struct hlist_head *);
1442                 new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1443                 new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1444                 if (!new_info_hash || !new_laddrhash) {
1445                         kvfree(new_info_hash);
1446                         kvfree(new_laddrhash);
1447                 } else {
1448                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1449                 }
1450                 if (!fib_info_hash_size)
1451                         goto failure;
1452         }
1453
1454         fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1455         if (!fi)
1456                 goto failure;
1457         fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1458                                               cfg->fc_mx_len, extack);
1459         if (IS_ERR(fi->fib_metrics)) {
1460                 err = PTR_ERR(fi->fib_metrics);
1461                 kfree(fi);
1462                 return ERR_PTR(err);
1463         }
1464
1465         fi->fib_net = net;
1466         fi->fib_protocol = cfg->fc_protocol;
1467         fi->fib_scope = cfg->fc_scope;
1468         fi->fib_flags = cfg->fc_flags;
1469         fi->fib_priority = cfg->fc_priority;
1470         fi->fib_prefsrc = cfg->fc_prefsrc;
1471         fi->fib_type = cfg->fc_type;
1472         fi->fib_tb_id = cfg->fc_table;
1473
1474         fi->fib_nhs = nhs;
1475         if (nh) {
1476                 if (!nexthop_get(nh)) {
1477                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1478                         err = -EINVAL;
1479                 } else {
1480                         err = 0;
1481                         fi->nh = nh;
1482                 }
1483         } else {
1484                 change_nexthops(fi) {
1485                         nexthop_nh->nh_parent = fi;
1486                 } endfor_nexthops(fi)
1487
1488                 if (cfg->fc_mp)
1489                         err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1490                                           extack);
1491                 else
1492                         err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1493         }
1494
1495         if (err != 0)
1496                 goto failure;
1497
1498         if (fib_props[cfg->fc_type].error) {
1499                 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1500                         NL_SET_ERR_MSG(extack,
1501                                        "Gateway, device and multipath can not be specified for this route type");
1502                         goto err_inval;
1503                 }
1504                 goto link_it;
1505         } else {
1506                 switch (cfg->fc_type) {
1507                 case RTN_UNICAST:
1508                 case RTN_LOCAL:
1509                 case RTN_BROADCAST:
1510                 case RTN_ANYCAST:
1511                 case RTN_MULTICAST:
1512                         break;
1513                 default:
1514                         NL_SET_ERR_MSG(extack, "Invalid route type");
1515                         goto err_inval;
1516                 }
1517         }
1518
1519         if (cfg->fc_scope > RT_SCOPE_HOST) {
1520                 NL_SET_ERR_MSG(extack, "Invalid scope");
1521                 goto err_inval;
1522         }
1523
1524         if (fi->nh) {
1525                 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1526                 if (err)
1527                         goto failure;
1528         } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1529                 struct fib_nh *nh = fi->fib_nh;
1530
1531                 /* Local address is added. */
1532                 if (nhs != 1) {
1533                         NL_SET_ERR_MSG(extack,
1534                                        "Route with host scope can not have multiple nexthops");
1535                         goto err_inval;
1536                 }
1537                 if (nh->fib_nh_gw_family) {
1538                         NL_SET_ERR_MSG(extack,
1539                                        "Route with host scope can not have a gateway");
1540                         goto err_inval;
1541                 }
1542                 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1543                 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1544                 err = -ENODEV;
1545                 if (!nh->fib_nh_dev)
1546                         goto failure;
1547                 netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1548                                      GFP_KERNEL);
1549         } else {
1550                 int linkdown = 0;
1551
1552                 change_nexthops(fi) {
1553                         err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1554                                            cfg->fc_table, cfg->fc_scope,
1555                                            extack);
1556                         if (err != 0)
1557                                 goto failure;
1558                         if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1559                                 linkdown++;
1560                 } endfor_nexthops(fi)
1561                 if (linkdown == fi->fib_nhs)
1562                         fi->fib_flags |= RTNH_F_LINKDOWN;
1563         }
1564
1565         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1566                 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1567                 goto err_inval;
1568         }
1569
1570         if (!fi->nh) {
1571                 change_nexthops(fi) {
1572                         fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1573                                                   fi->fib_scope);
1574                         if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1575                                 fi->fib_nh_is_v6 = true;
1576                 } endfor_nexthops(fi)
1577
1578                 fib_rebalance(fi);
1579         }
1580
1581 link_it:
1582         ofi = fib_find_info(fi);
1583         if (ofi) {
1584                 fi->fib_dead = 1;
1585                 free_fib_info(fi);
1586                 refcount_inc(&ofi->fib_treeref);
1587                 return ofi;
1588         }
1589
1590         refcount_set(&fi->fib_treeref, 1);
1591         refcount_set(&fi->fib_clntref, 1);
1592         spin_lock_bh(&fib_info_lock);
1593         fib_info_cnt++;
1594         hlist_add_head(&fi->fib_hash,
1595                        &fib_info_hash[fib_info_hashfn(fi)]);
1596         if (fi->fib_prefsrc) {
1597                 struct hlist_head *head;
1598
1599                 head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1600                 hlist_add_head(&fi->fib_lhash, head);
1601         }
1602         if (fi->nh) {
1603                 list_add(&fi->nh_list, &nh->fi_list);
1604         } else {
1605                 change_nexthops(fi) {
1606                         struct hlist_head *head;
1607
1608                         if (!nexthop_nh->fib_nh_dev)
1609                                 continue;
1610                         head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1611                         hlist_add_head(&nexthop_nh->nh_hash, head);
1612                 } endfor_nexthops(fi)
1613         }
1614         spin_unlock_bh(&fib_info_lock);
1615         return fi;
1616
1617 err_inval:
1618         err = -EINVAL;
1619
1620 failure:
1621         if (fi) {
1622                 fi->fib_dead = 1;
1623                 free_fib_info(fi);
1624         }
1625
1626         return ERR_PTR(err);
1627 }
1628
1629 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1630                      u8 rt_family, unsigned char *flags, bool skip_oif)
1631 {
1632         if (nhc->nhc_flags & RTNH_F_DEAD)
1633                 *flags |= RTNH_F_DEAD;
1634
1635         if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1636                 *flags |= RTNH_F_LINKDOWN;
1637
1638                 rcu_read_lock();
1639                 switch (nhc->nhc_family) {
1640                 case AF_INET:
1641                         if (ip_ignore_linkdown(nhc->nhc_dev))
1642                                 *flags |= RTNH_F_DEAD;
1643                         break;
1644                 case AF_INET6:
1645                         if (ip6_ignore_linkdown(nhc->nhc_dev))
1646                                 *flags |= RTNH_F_DEAD;
1647                         break;
1648                 }
1649                 rcu_read_unlock();
1650         }
1651
1652         switch (nhc->nhc_gw_family) {
1653         case AF_INET:
1654                 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1655                         goto nla_put_failure;
1656                 break;
1657         case AF_INET6:
1658                 /* if gateway family does not match nexthop family
1659                  * gateway is encoded as RTA_VIA
1660                  */
1661                 if (rt_family != nhc->nhc_gw_family) {
1662                         int alen = sizeof(struct in6_addr);
1663                         struct nlattr *nla;
1664                         struct rtvia *via;
1665
1666                         nla = nla_reserve(skb, RTA_VIA, alen + 2);
1667                         if (!nla)
1668                                 goto nla_put_failure;
1669
1670                         via = nla_data(nla);
1671                         via->rtvia_family = AF_INET6;
1672                         memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1673                 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1674                                             &nhc->nhc_gw.ipv6) < 0) {
1675                         goto nla_put_failure;
1676                 }
1677                 break;
1678         }
1679
1680         *flags |= (nhc->nhc_flags &
1681                    (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1682
1683         if (!skip_oif && nhc->nhc_dev &&
1684             nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1685                 goto nla_put_failure;
1686
1687         if (nhc->nhc_lwtstate &&
1688             lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1689                                 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1690                 goto nla_put_failure;
1691
1692         return 0;
1693
1694 nla_put_failure:
1695         return -EMSGSIZE;
1696 }
1697 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1698
1699 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1700 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1701                     int nh_weight, u8 rt_family, u32 nh_tclassid)
1702 {
1703         const struct net_device *dev = nhc->nhc_dev;
1704         struct rtnexthop *rtnh;
1705         unsigned char flags = 0;
1706
1707         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1708         if (!rtnh)
1709                 goto nla_put_failure;
1710
1711         rtnh->rtnh_hops = nh_weight - 1;
1712         rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1713
1714         if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1715                 goto nla_put_failure;
1716
1717         rtnh->rtnh_flags = flags;
1718
1719         if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1720                 goto nla_put_failure;
1721
1722         /* length of rtnetlink header + attributes */
1723         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1724
1725         return 0;
1726
1727 nla_put_failure:
1728         return -EMSGSIZE;
1729 }
1730 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1731 #endif
1732
1733 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1734 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1735 {
1736         struct nlattr *mp;
1737
1738         mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1739         if (!mp)
1740                 goto nla_put_failure;
1741
1742         if (unlikely(fi->nh)) {
1743                 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1744                         goto nla_put_failure;
1745                 goto mp_end;
1746         }
1747
1748         for_nexthops(fi) {
1749                 u32 nh_tclassid = 0;
1750 #ifdef CONFIG_IP_ROUTE_CLASSID
1751                 nh_tclassid = nh->nh_tclassid;
1752 #endif
1753                 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1754                                     AF_INET, nh_tclassid) < 0)
1755                         goto nla_put_failure;
1756         } endfor_nexthops(fi);
1757
1758 mp_end:
1759         nla_nest_end(skb, mp);
1760
1761         return 0;
1762
1763 nla_put_failure:
1764         return -EMSGSIZE;
1765 }
1766 #else
1767 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1768 {
1769         return 0;
1770 }
1771 #endif
1772
1773 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1774                   const struct fib_rt_info *fri, unsigned int flags)
1775 {
1776         unsigned int nhs = fib_info_num_path(fri->fi);
1777         struct fib_info *fi = fri->fi;
1778         u32 tb_id = fri->tb_id;
1779         struct nlmsghdr *nlh;
1780         struct rtmsg *rtm;
1781
1782         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1783         if (!nlh)
1784                 return -EMSGSIZE;
1785
1786         rtm = nlmsg_data(nlh);
1787         rtm->rtm_family = AF_INET;
1788         rtm->rtm_dst_len = fri->dst_len;
1789         rtm->rtm_src_len = 0;
1790         rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1791         if (tb_id < 256)
1792                 rtm->rtm_table = tb_id;
1793         else
1794                 rtm->rtm_table = RT_TABLE_COMPAT;
1795         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1796                 goto nla_put_failure;
1797         rtm->rtm_type = fri->type;
1798         rtm->rtm_flags = fi->fib_flags;
1799         rtm->rtm_scope = fi->fib_scope;
1800         rtm->rtm_protocol = fi->fib_protocol;
1801
1802         if (rtm->rtm_dst_len &&
1803             nla_put_in_addr(skb, RTA_DST, fri->dst))
1804                 goto nla_put_failure;
1805         if (fi->fib_priority &&
1806             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1807                 goto nla_put_failure;
1808         if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1809                 goto nla_put_failure;
1810
1811         if (fi->fib_prefsrc &&
1812             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1813                 goto nla_put_failure;
1814
1815         if (fi->nh) {
1816                 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1817                         goto nla_put_failure;
1818                 if (nexthop_is_blackhole(fi->nh))
1819                         rtm->rtm_type = RTN_BLACKHOLE;
1820                 if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode))
1821                         goto offload;
1822         }
1823
1824         if (nhs == 1) {
1825                 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1826                 unsigned char flags = 0;
1827
1828                 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1829                         goto nla_put_failure;
1830
1831                 rtm->rtm_flags = flags;
1832 #ifdef CONFIG_IP_ROUTE_CLASSID
1833                 if (nhc->nhc_family == AF_INET) {
1834                         struct fib_nh *nh;
1835
1836                         nh = container_of(nhc, struct fib_nh, nh_common);
1837                         if (nh->nh_tclassid &&
1838                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1839                                 goto nla_put_failure;
1840                 }
1841 #endif
1842         } else {
1843                 if (fib_add_multipath(skb, fi) < 0)
1844                         goto nla_put_failure;
1845         }
1846
1847 offload:
1848         if (fri->offload)
1849                 rtm->rtm_flags |= RTM_F_OFFLOAD;
1850         if (fri->trap)
1851                 rtm->rtm_flags |= RTM_F_TRAP;
1852         if (fri->offload_failed)
1853                 rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1854
1855         nlmsg_end(skb, nlh);
1856         return 0;
1857
1858 nla_put_failure:
1859         nlmsg_cancel(skb, nlh);
1860         return -EMSGSIZE;
1861 }
1862
1863 /*
1864  * Update FIB if:
1865  * - local address disappeared -> we must delete all the entries
1866  *   referring to it.
1867  * - device went down -> we must shutdown all nexthops going via it.
1868  */
1869 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1870 {
1871         int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1872         struct net *net = dev_net(dev);
1873         struct hlist_head *head;
1874         struct fib_info *fi;
1875         int ret = 0;
1876
1877         if (!fib_info_laddrhash || local == 0)
1878                 return 0;
1879
1880         head = fib_info_laddrhash_bucket(net, local);
1881         hlist_for_each_entry(fi, head, fib_lhash) {
1882                 if (!net_eq(fi->fib_net, net) ||
1883                     fi->fib_tb_id != tb_id)
1884                         continue;
1885                 if (fi->fib_prefsrc == local) {
1886                         fi->fib_flags |= RTNH_F_DEAD;
1887                         ret++;
1888                 }
1889         }
1890         return ret;
1891 }
1892
1893 static int call_fib_nh_notifiers(struct fib_nh *nh,
1894                                  enum fib_event_type event_type)
1895 {
1896         bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1897         struct fib_nh_notifier_info info = {
1898                 .fib_nh = nh,
1899         };
1900
1901         switch (event_type) {
1902         case FIB_EVENT_NH_ADD:
1903                 if (nh->fib_nh_flags & RTNH_F_DEAD)
1904                         break;
1905                 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1906                         break;
1907                 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1908                                            &info.info);
1909         case FIB_EVENT_NH_DEL:
1910                 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1911                     (nh->fib_nh_flags & RTNH_F_DEAD))
1912                         return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1913                                                    event_type, &info.info);
1914                 break;
1915         default:
1916                 break;
1917         }
1918
1919         return NOTIFY_DONE;
1920 }
1921
1922 /* Update the PMTU of exceptions when:
1923  * - the new MTU of the first hop becomes smaller than the PMTU
1924  * - the old MTU was the same as the PMTU, and it limited discovery of
1925  *   larger MTUs on the path. With that limit raised, we can now
1926  *   discover larger MTUs
1927  * A special case is locked exceptions, for which the PMTU is smaller
1928  * than the minimal accepted PMTU:
1929  * - if the new MTU is greater than the PMTU, don't make any change
1930  * - otherwise, unlock and set PMTU
1931  */
1932 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1933 {
1934         struct fnhe_hash_bucket *bucket;
1935         int i;
1936
1937         bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1938         if (!bucket)
1939                 return;
1940
1941         for (i = 0; i < FNHE_HASH_SIZE; i++) {
1942                 struct fib_nh_exception *fnhe;
1943
1944                 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1945                      fnhe;
1946                      fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1947                         if (fnhe->fnhe_mtu_locked) {
1948                                 if (new <= fnhe->fnhe_pmtu) {
1949                                         fnhe->fnhe_pmtu = new;
1950                                         fnhe->fnhe_mtu_locked = false;
1951                                 }
1952                         } else if (new < fnhe->fnhe_pmtu ||
1953                                    orig == fnhe->fnhe_pmtu) {
1954                                 fnhe->fnhe_pmtu = new;
1955                         }
1956                 }
1957         }
1958 }
1959
1960 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1961 {
1962         struct hlist_head *head = fib_info_devhash_bucket(dev);
1963         struct fib_nh *nh;
1964
1965         hlist_for_each_entry(nh, head, nh_hash) {
1966                 if (nh->fib_nh_dev == dev)
1967                         fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1968         }
1969 }
1970
1971 /* Event              force Flags           Description
1972  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1973  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1974  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1975  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1976  *
1977  * only used when fib_nh is built into fib_info
1978  */
1979 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1980 {
1981         struct hlist_head *head = fib_info_devhash_bucket(dev);
1982         struct fib_info *prev_fi = NULL;
1983         int scope = RT_SCOPE_NOWHERE;
1984         struct fib_nh *nh;
1985         int ret = 0;
1986
1987         if (force)
1988                 scope = -1;
1989
1990         hlist_for_each_entry(nh, head, nh_hash) {
1991                 struct fib_info *fi = nh->nh_parent;
1992                 int dead;
1993
1994                 BUG_ON(!fi->fib_nhs);
1995                 if (nh->fib_nh_dev != dev || fi == prev_fi)
1996                         continue;
1997                 prev_fi = fi;
1998                 dead = 0;
1999                 change_nexthops(fi) {
2000                         if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
2001                                 dead++;
2002                         else if (nexthop_nh->fib_nh_dev == dev &&
2003                                  nexthop_nh->fib_nh_scope != scope) {
2004                                 switch (event) {
2005                                 case NETDEV_DOWN:
2006                                 case NETDEV_UNREGISTER:
2007                                         nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2008                                         fallthrough;
2009                                 case NETDEV_CHANGE:
2010                                         nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2011                                         break;
2012                                 }
2013                                 call_fib_nh_notifiers(nexthop_nh,
2014                                                       FIB_EVENT_NH_DEL);
2015                                 dead++;
2016                         }
2017 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2018                         if (event == NETDEV_UNREGISTER &&
2019                             nexthop_nh->fib_nh_dev == dev) {
2020                                 dead = fi->fib_nhs;
2021                                 break;
2022                         }
2023 #endif
2024                 } endfor_nexthops(fi)
2025                 if (dead == fi->fib_nhs) {
2026                         switch (event) {
2027                         case NETDEV_DOWN:
2028                         case NETDEV_UNREGISTER:
2029                                 fi->fib_flags |= RTNH_F_DEAD;
2030                                 fallthrough;
2031                         case NETDEV_CHANGE:
2032                                 fi->fib_flags |= RTNH_F_LINKDOWN;
2033                                 break;
2034                         }
2035                         ret++;
2036                 }
2037
2038                 fib_rebalance(fi);
2039         }
2040
2041         return ret;
2042 }
2043
2044 /* Must be invoked inside of an RCU protected region.  */
2045 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2046 {
2047         struct fib_info *fi = NULL, *last_resort = NULL;
2048         struct hlist_head *fa_head = res->fa_head;
2049         struct fib_table *tb = res->table;
2050         u8 slen = 32 - res->prefixlen;
2051         int order = -1, last_idx = -1;
2052         struct fib_alias *fa, *fa1 = NULL;
2053         u32 last_prio = res->fi->fib_priority;
2054         dscp_t last_dscp = 0;
2055
2056         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2057                 struct fib_info *next_fi = fa->fa_info;
2058                 struct fib_nh_common *nhc;
2059
2060                 if (fa->fa_slen != slen)
2061                         continue;
2062                 if (fa->fa_dscp &&
2063                     fa->fa_dscp != inet_dsfield_to_dscp(flp->flowi4_tos))
2064                         continue;
2065                 if (fa->tb_id != tb->tb_id)
2066                         continue;
2067                 if (next_fi->fib_priority > last_prio &&
2068                     fa->fa_dscp == last_dscp) {
2069                         if (last_dscp)
2070                                 continue;
2071                         break;
2072                 }
2073                 if (next_fi->fib_flags & RTNH_F_DEAD)
2074                         continue;
2075                 last_dscp = fa->fa_dscp;
2076                 last_prio = next_fi->fib_priority;
2077
2078                 if (next_fi->fib_scope != res->scope ||
2079                     fa->fa_type != RTN_UNICAST)
2080                         continue;
2081
2082                 nhc = fib_info_nhc(next_fi, 0);
2083                 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2084                         continue;
2085
2086                 fib_alias_accessed(fa);
2087
2088                 if (!fi) {
2089                         if (next_fi != res->fi)
2090                                 break;
2091                         fa1 = fa;
2092                 } else if (!fib_detect_death(fi, order, &last_resort,
2093                                              &last_idx, fa1->fa_default)) {
2094                         fib_result_assign(res, fi);
2095                         fa1->fa_default = order;
2096                         goto out;
2097                 }
2098                 fi = next_fi;
2099                 order++;
2100         }
2101
2102         if (order <= 0 || !fi) {
2103                 if (fa1)
2104                         fa1->fa_default = -1;
2105                 goto out;
2106         }
2107
2108         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2109                               fa1->fa_default)) {
2110                 fib_result_assign(res, fi);
2111                 fa1->fa_default = order;
2112                 goto out;
2113         }
2114
2115         if (last_idx >= 0)
2116                 fib_result_assign(res, last_resort);
2117         fa1->fa_default = last_idx;
2118 out:
2119         return;
2120 }
2121
2122 /*
2123  * Dead device goes up. We wake up dead nexthops.
2124  * It takes sense only on multipath routes.
2125  *
2126  * only used when fib_nh is built into fib_info
2127  */
2128 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2129 {
2130         struct fib_info *prev_fi;
2131         struct hlist_head *head;
2132         struct fib_nh *nh;
2133         int ret;
2134
2135         if (!(dev->flags & IFF_UP))
2136                 return 0;
2137
2138         if (nh_flags & RTNH_F_DEAD) {
2139                 unsigned int flags = dev_get_flags(dev);
2140
2141                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2142                         nh_flags |= RTNH_F_LINKDOWN;
2143         }
2144
2145         prev_fi = NULL;
2146         head = fib_info_devhash_bucket(dev);
2147         ret = 0;
2148
2149         hlist_for_each_entry(nh, head, nh_hash) {
2150                 struct fib_info *fi = nh->nh_parent;
2151                 int alive;
2152
2153                 BUG_ON(!fi->fib_nhs);
2154                 if (nh->fib_nh_dev != dev || fi == prev_fi)
2155                         continue;
2156
2157                 prev_fi = fi;
2158                 alive = 0;
2159                 change_nexthops(fi) {
2160                         if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2161                                 alive++;
2162                                 continue;
2163                         }
2164                         if (!nexthop_nh->fib_nh_dev ||
2165                             !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2166                                 continue;
2167                         if (nexthop_nh->fib_nh_dev != dev ||
2168                             !__in_dev_get_rtnl(dev))
2169                                 continue;
2170                         alive++;
2171                         nexthop_nh->fib_nh_flags &= ~nh_flags;
2172                         call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2173                 } endfor_nexthops(fi)
2174
2175                 if (alive > 0) {
2176                         fi->fib_flags &= ~nh_flags;
2177                         ret++;
2178                 }
2179
2180                 fib_rebalance(fi);
2181         }
2182
2183         return ret;
2184 }
2185
2186 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2187 static bool fib_good_nh(const struct fib_nh *nh)
2188 {
2189         int state = NUD_REACHABLE;
2190
2191         if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2192                 struct neighbour *n;
2193
2194                 rcu_read_lock();
2195
2196                 if (likely(nh->fib_nh_gw_family == AF_INET))
2197                         n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2198                                                    (__force u32)nh->fib_nh_gw4);
2199                 else if (nh->fib_nh_gw_family == AF_INET6)
2200                         n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2201                                                            &nh->fib_nh_gw6);
2202                 else
2203                         n = NULL;
2204                 if (n)
2205                         state = READ_ONCE(n->nud_state);
2206
2207                 rcu_read_unlock();
2208         }
2209
2210         return !!(state & NUD_VALID);
2211 }
2212
2213 void fib_select_multipath(struct fib_result *res, int hash)
2214 {
2215         struct fib_info *fi = res->fi;
2216         struct net *net = fi->fib_net;
2217         bool first = false;
2218
2219         if (unlikely(res->fi->nh)) {
2220                 nexthop_path_fib_result(res, hash);
2221                 return;
2222         }
2223
2224         change_nexthops(fi) {
2225                 if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) {
2226                         if (!fib_good_nh(nexthop_nh))
2227                                 continue;
2228                         if (!first) {
2229                                 res->nh_sel = nhsel;
2230                                 res->nhc = &nexthop_nh->nh_common;
2231                                 first = true;
2232                         }
2233                 }
2234
2235                 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2236                         continue;
2237
2238                 res->nh_sel = nhsel;
2239                 res->nhc = &nexthop_nh->nh_common;
2240                 return;
2241         } endfor_nexthops(fi);
2242 }
2243 #endif
2244
2245 void fib_select_path(struct net *net, struct fib_result *res,
2246                      struct flowi4 *fl4, const struct sk_buff *skb)
2247 {
2248         if (fl4->flowi4_oif)
2249                 goto check_saddr;
2250
2251 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2252         if (fib_info_num_path(res->fi) > 1) {
2253                 int h = fib_multipath_hash(net, fl4, skb, NULL);
2254
2255                 fib_select_multipath(res, h);
2256         }
2257         else
2258 #endif
2259         if (!res->prefixlen &&
2260             res->table->tb_num_default > 1 &&
2261             res->type == RTN_UNICAST)
2262                 fib_select_default(fl4, res);
2263
2264 check_saddr:
2265         if (!fl4->saddr)
2266                 fl4->saddr = fib_result_prefsrc(net, res);
2267 }